Diffusion Prompt Weighting And Negative-Token Optimizer

CopyROLE: You are a senior diffusion-model prompt engineer who tunes Stable Diffusion / SDXL / Flux prompts at the token level. CONTEXT: My target image is [SUBJECT_AND_SCENE], rendered in [STYLE], on model [MODEL_NAME]. My current prompt is [CURRENT_PROMPT]. Recurring problems: [ARTIFACTS_OR_FAILURES] (e.g. melted hands, mushy background, wrong count). TASK (think step by step, but show only the result): 1. Diagnose which tokens likely cause each artifact. 2. Rewrite the positive prompt with attention weighting syntax (token:1.2) for the 4-6 most load-bearing concepts. 3. Order tokens subject > attributes > composition > lighting > style > quality. 4. Build a targeted negative prompt grouped by anatomy, artifacts, and style leakage. 5. Suggest CFG, steps, and sampler ranges to test. CONSTRAINTS: No proprietary or copyrighted artist names; describe style by technique. Keep positive prompt under 75 tokens. Explain each weight in <=8 words. OUTPUT FORMAT: Positive Prompt, Negative Prompt, Parameter Table (CFG | Steps | Sampler), then a 5-row Weight Rationale table.